585-paper-4

8/12/2019 585-paper-4

1/17

The Quarterly Review of Economics and Finance 51 (2011) 88104

Contents lists available atScienceDirect

The Quarterly Review of Economics and Finance

j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / q r e f

Financial development and economic growth: New evidence from panel data

M. Kabir Hassan a,, Benito Sanchez b, Jung-Suk Yu c

a Department of Economics and Finance, University of New Orleans, 2000 Lakeshore Drive, New Orleans, LA 70148, USAb Department of Economics and Finance, Kean University, Union, NJ 07083, USAc School of Urban Planning & Real Estate Studies, Dankook University, Gyeonggi-do, 448-701, Republic of Korea

a r t i c l e i n f o

Article history:

Received 8 September 2009

Received in revised form 18 August 2010

Accepted 20 September 2010

Available online 29 September 2010

JEL classification:

G21

O16

C33

Keywords:

Financial development

Economic growth

Panel regression

Granger causality tests

a b s t r a c t

This study provides evidence on the role of financial development in accounting for economic growthin low- and middle-income countries classified by geographic regions. To document the relationship

between financial development and economic growth, we estimate both panel regressions and variance

decompositionsof annual GDPper capita growthratesto examinewhat proxymeasures of financialdevel-

opmentare most important in accounting for economic growthover time andhow much they contribute

to explaining economic growth across geographicregions and income groups. We find a positive relation-

shipbetweenfinancial development andeconomicgrowth in developing countries.Moreover,short-term

multivariate analysis provides mixed results: a two-way causality relationship between finance and

growth for most regions and one-way causality from growth to finance for the two poorest regions.

Furthermore, other variables from the real sector such as trade and government expenditure play an

important role in explaining economic growth. Therefore, it seems that a well-functioning financial sys-

tem is a necessary but not sufficient condition to reach steady economic growth in developing countries.

Published by Elsevier B.V. on behalf of The Board of Trustees of the University of Illinois.

1. Introduction

The relationship between financial development and economic

growth has received a great deal of attention in recent decades.

However, there are conflicting views concerning the role that the

financial system plays in economic growth. For example, while

Levine (1997)believes that financial intermediaries enhance eco-

nomicefficiency,and ultimately growth, by helping allocate capital

to its best uses, Lucas (1988)asserts that the role of the financial

sector in economic growth is over-stressed. Notwithstanding the

controversy, modern theoretical literature on the financegrowth

nexus combines the endogenous growth theory and microeco-

nomics of financial systems (Grossman & Helpman, 1991; Khan,

2001; Lucas, 1988; Pagano, 1993; Rebelo, 1991; Romer, 1986;

among others).

Early studies on financial development (FD) and economic

growth (EG) were based on cross-country analysis. For instance,

Goldsmith (1969), King and Levine (1993a, 1993b), and Levine and

Zervos (1998) used cross-country analysis to studythe relationship

between financial development and economic growth. While their

Corresponding author. Tel.: +1 504 280 6163; fax: +1 504 280 6397.

E-mail addresses: [email protected](M.K. Hassan), [email protected]

(B. Sanchez),[email protected](J.-S. Yu).

findings suggest that finance helps to predict growth, these studies

do not deal formally with the issue of causality, nor do they exploit

the time-series properties of the data.1 Furthermore, conclusions

based on cross-country analysis are sensitive to the sample coun-

tries, estimation methods, data frequency, functional form of the

relationship, and proxy measures chosen in the study, all of which

raise doubts about the reliability of cross-country regression anal-

ysis (seeAl-Awad & Harb, 2005; Chuah & Thai, 2004; Hassan &

Bashir, 2003; Khan & Senhadji, 2003).

Panel time-series analysis, on the other hand, exploits time-

series and cross-sectional variations in data and avoids biases

associated with cross-sectional regressions by taking the country-

specific fixed effect into account (Levine, 2005). To mitigate the

shortcomings of cross-sectional analysis, this paper examines the

dynamic relationship between economic growth and financial

development across geographic regions and income groups using

time-series analysis.

In retrospect, our interest was motivated by three factors. First,

it is argued that well-developed domestic financial sectors, such

as those of developed countries [high-income Organization and

Economic Cooperation and Development (OECD) countries], can

1 However,thesestudiesdefinecontrolvariablesand measuresof financialdevel-

opment that are typically used in time-series analysis.

1062-9769/$ see front matter. Published by Elsevier B.V. on behalf of The Board of Trustees of the University of Illinois.

doi:10.1016/j.qref.2010.09.001
http://localhost/var/www/apps/conversion/tmp/scratch_9/dx.doi.org/10.1016/j.qref.2010.09.001http://localhost/var/www/apps/conversion/tmp/scratch_9/dx.doi.org/10.1016/j.qref.2010.09.001http://www.sciencedirect.com/science/journal/10629769http://www.elsevier.com/locate/qrefmailto:[email protected]:[email protected]:[email protected]://localhost/var/www/apps/conversion/tmp/scratch_9/dx.doi.org/10.1016/j.qref.2010.09.001http://localhost/var/www/apps/conversion/tmp/scratch_9/dx.doi.org/10.1016/j.qref.2010.09.001mailto:[email protected]:[email protected]:[email protected]://www.elsevier.com/locate/qrefhttp://www.sciencedirect.com/science/journal/10629769http://localhost/var/www/apps/conversion/tmp/scratch_9/dx.doi.org/10.1016/j.qref.2010.09.001

8/12/2019 585-paper-4

2/17

M.K. Hassan et al. / The Quarterly Review of Economics and Finance 51 (2011) 88104 89

significantly contribute to an increase in savings and investment

rate and, eventually lead to economic growth(Becsi & Wang, 1997).

Following this premise, most developing countries have reformed

their economic and financial systems to improve the efficiency

of their financial intermediaries with the objective of achieving

financial sector development and promoting growth, starting in

the 1980s. Therefore, we document the progress achieved by these

countries over the last three decades in terms of revamping their

financial systems, and assess the links between the reforms and

economic performance.

Second, we employ unbalanced panel estimations and various

multivariate time-series analysis technique to establish the direc-

tion, timing, and strength of the causal link between the real and

financial sectors across geographic regions and income groups so

that we may explore some policy implications. We also use finan-

cial development indicators employed in the literature and draw

some conclusions about their impact on economic growth as mea-

sured by the annual growth rate of the domestic product (GDP) per

capita.

Finally, instead of using heterogeneous cross-country samples,

we investigate different geographic regions, each of which has

a relatively homogeneous sample of countries. This is adequate

for assessing the links between economic growth and financial

development. Most time-series studies have analyzed either het-

erogeneous countries or a set of stand-alone countries.2 We take

a different approach in this paper. Rather than pooling worldwide

data or analyzing each country, we study the relationship between

finance and growth in geographic regions using World Bank clas-

sifications. The World Bank only categorizes geographic regions

as low- or middle-income countries. High-income countries are

excludedin itsclassificationof geographicregions.Therefore, coun-

tries in each geographic region are homogenous with respect to

the level of GDP per capita, financial development, and culture.

Furthermore, we are able to capture the temporal dimension of

the economic reforms by combining time-series with geographic

cross-sectional data. The main advantage of this approach is that

we areableto useenough data toestimateparametersin panel dataregression and other multivariate analysis techniques that other-

wise could notbe estimated fora single country, and yetdocument

the financegrowth associationwith the objective of deriving some

policy implications for each region (andthe countries thatbelong to

the region). Also, to benchmark middle- and low-income countries

against high-income countries, we include high-income countries

classifiedby theWorldBank as either high-income OECD countries

or high-income non-OECD countries.

Using a neo-classical growth model, and in agreement with

King and Levine (1993a), and Levine, Loayza, and Beck (2000),

among others, we find strong long-run linkages between finan-

cial development and economic growth for developing countries.

Specifically, as predicted in neo-classical growth models (Pagano,

1993), domestic grosssavings is positivelyrelated to growth. More-over, other proxies for financial development, such as domestic

creditprovidedby thebanking sectorand domestic creditprovided

to the private sector, are positively related to economic growth.

Furthermore, consistent with the standard results for condi-

tional convergence (Barro, 1997; Bekaert et al., 2005),we find that

a low initial GDP per capita level is associated with a higher-rate

of economic growth for most regions, after controlling for financial

variables and real sector variables.

2 For instance, Calderonand Liu(2003) and Bekaert,Harvey, andLundblad (2005)

ran regressions using 109 and 95 worldwide countries, respectively. Shan, Morris,

and Sun (2001)studied 10 developed countries running regressions for each coun-

try.

Likewise, using the Granger causality test developed by Toda

and Yamamoto (1995), we find a two-way causality between

finance and growth in all regions but Sub-Saharan Africa and East

Asia & Pacific. This result is consistent with Shan, Morris, and

Sun (2001)and Demetriades and Hussein (1996), who found bi-

directional causality between finance and growth, but contrary to

Christopoulos and Tsionas (2004),who found that the direction is

from finance to growth. The results also provide some support to

the theoretical models ofBlackburn and Huang (1998)andKhan

(2001), which predict a two-way causality between finance and

growth.

However, we findthat thecausality runs from growthto finance

in South Asia and in Sub-Saharan Africa, the two poorest regions

in our sample. This result supports the views ofGurley and Shaw

(1967), Goldsmith (1969), and Jung (1986), who hypothesized

that in developing countries, growth leads finance because of the

increasing demand for financial services.

The paper is organizedas follows. Section 2 provides a literature

review. Section3 describes the data and the proxy measures of

financial development, real sector, and economic growth. Section 4

describes the unbalancedpanel estimations and multivariate time-

series methodologies applied in the paper. Section 5analyzes the

empirical results, and Section6provides conclusions.

2. Literature review

Since the pioneering contributions of Goldsmith (1969),

McKinnon (1973),and Shaw (1973)on the role of FD in promot-

ing EG, the relationship between EG and FD has remained an

important issue of debate among academics and policymakers (De

Gregorio & Guidotti, 1995).Early economic growth theory argued

that economic development is a process of innovations whereby

the interactions of innovations in both the financial and real sec-

tors provide a driving force for dynamic economic growth. In other

words, exogenous technological progress determines the long-run

growth rate, while financial intermediaries are not explicitly mod-eled to affect the long-run growth rate.

However, a growing contemporary theoretical and empirical

body of literature shows how financial intermediation mobilizes

savings, allocates resources, diversifies risks, and contributes to

economic growth (Greenwood & Jovanovic, 1990; Jbili, Enders,

& Treichel, 1997). The new growth theory argues that finan-

cial intermediaries and markets appear endogenously in response

to market incompleteness and, hence, contribute to long-term

growth. Financial institutions and markets, which arise endoge-

nously to mitigate the effects of information and transaction cost

frictions, influences decisions to invest in productivity-enhancing

activities through evaluating prospective entrepreneurs and fund-

ing the most promising ones. The underlying assumption is that

financial intermediaries can provide these evaluation and moni-toring services more efficiently than individuals.

An important set of authors in the literature agrees that there

is a relation between finance and economic growth. However,

they disagree about the direction of causality. On one hand, some

authors have theoretically and empirically shown that there is

causal direction from FD to EG. That is, policies that move toward

the development of financial systems lead to economic growth.

McKinnon (1973), King and Levine (1993a), Levineet al.(2000), and

Christopoulos and Tsionas (2004)support this argument. On the

other hand,otherauthors argue that thedirection is from economic

growth to financial development. Since the economy is growing,

there is an increasing demand forfinancial services that induces an

expansion in the financial sector. This view is supported by Gurley

and Shaw (1967),Goldsmith (1969),andJung (1986).

8/12/2019 585-paper-4

3/17

90 M.K. Hassan et al. / The Quarterly Review of Economics and Finance 51 (2011) 88104

Other authors argue that thecausal directionis two-way. Finan-

cial development (FD) and economic growth (EG) reinforce each

other. FD supports EG and EG renders support to FD. Patrick

(1966) postulatedthe stageof development hypothesis. At the early

stage, causality runs from finance to growth, but at later stages

causality runs from growth to finance. In the early stage of eco-

nomic development, finance causes growth by inducing real per

capita capital formation. Later on, the economy is in the growth

stage and there will be increasing demand for financial services,

which induces an expansion in the financial sector as well as

the real sector. This implies causality from growth to finance.

Blackburn and Huang (1998)also established a positive two-way

causal relationship between growth and financial development.

According to their analysis, private informed agents obtain exter-

nal financing for their projects through incentive-compatible loan

contracts, which are enforced through costly monitoring activi-

ties that lenders may delegate to financial intermediaries. More

recently, Khan (2001) also established a positive two-way causality

between finance and growth. He postulated that when borrowing

is limited, producers with access to loans from financial interme-

diaries obtain higher returns, which creates an incentive for others

to undertake the technology necessary to access investment loans,

which in turn reduces financing costs and increases economic

growth.

Levine (1997, 2005) surveyed a large amount of empirical

research that deals with the relationship between the financial

sector and long-run growth. Levine (1997)argued that financial

systems can accomplish five functions to ameliorate information

and transactions frictions and contributeto long-run growth. These

functions are: facilitating risk amelioration, acquiring information

about investments and allocating resources, monitoring managers

and exerting corporate control, mobilizing savings, and facilitating

exchange. These functions facilitate investment and, hence, higher

economic growth.

The results in the literature, however, are ambiguous. On one

hand, cross-country and panel data studies find a positive effect

of financial depth on economic growth after accounting for otherdeterminants of growth and potentialbiases induced by simultane-

ity, omitted variables or country-specific effects (Levine, 2005),

suggesting that the causality runs from finance to growth (see

Christopoulos & Tsionas, 2004; Khan & Senhadji, 2003; King &

Levine, 1993a, 1993b; Levine et al., 2000).Furthermore,Claessens

and Laeven (2005) related banking competition and industrial

growth and found that the higher the competition among banks,

the faster the growth of finance-dependent industries, suggest-

ing also that higher financial development precedes economic

growth.

On the other hand,Demetriades and Hussein (1996)andShan

et al. (2001), using time-series techniques, found that the causality

is bi-directional for the majority of countries in their sample. Fur-

thermore, Luintel and Khan(1999), using a sampleof 10developingcountries, concluded that the causality between financial develop-

ment and output growth is bi-directional for the 10 countries they

studied. Calderon and Liu, using a sample of 109 developing and

developed countries, found evidence that financial development

generally leads to economic growth for developed countries, but

that the Granger causality is two-way for developing countries.

Since financial development is not easily measurable, papers

attempting to study the link between financial deepening and

growthhave chosena numberof proxy measures andsubsequently

have come up with different results (Al-Awad & Harb, 2005; Chuah

& Thai, 2004; Hassan& Bashir, 2003; Khan & Senhadji, 2003;King &

Levine, 1993a; Savvides, 1995;among others). However, the gen-

eral consensus of these studies is that there is a positive correlation

between financial development and economic growth.

3. Data and proxy measures

3.1. Structuring the panel dataset

Our sample period is 19802007, which covers an era of finan-

cial liberalization and development in many countries as well

as output expansion, money growth, and an increasing volume

of investment. Our comprehensive original dataset includes 168

countries and uses the nested panel data structure from the World

BanksWorld Development Indicators(WDI) 2009 database.3

To study how financial development and the real sector are

linked to economic growth, we followed the World Bank classifi-

cations, which categorize all World Bank member economies, and

all other economies with populations of more than 30,000 people,

into six geographic regions and four income groups. Countries by

regions, variable definitions, and time-series averages are listed in

Appendix A.

This dataset allows us effectively to estimate panel regressions

and analyze various multivariate time-series models within each

geographic region and income group. Despite the shortcomings

from regional aggregations, we believe that our approach to esti-

mate models based on geographic regions and income groups

has several advantages in terms of providing policy implications

compared to previous time-series and cross-sectional studies that

include large numbers of heterogeneous countries or individual

cases. Each region is a set of homogeneous countries (e.g., similar

GDP per capita, finance structure, culture, etc.) but with enough

variation in explanatory variables to perform panel regressions

and multivariate time-series models. Therefore, it is possible to

document theassociation between finance andgrowth by dynami-

cally examining different economic roles, causality, directions, and

timing among proxy measures for financial development and eco-

nomic growth across geographic regions and income groups with

the objective of documenting financial liberalization and assessing

some policy implications.

3.2. Proxy measures for financial development and economic

growth

Various measures have been used in the literature to proxy for

the level of financial development, ranging from interest rates,

to monetary aggregates, to the ratio of the size of the banking sys-

tem to GDP (Al-Awad & Harb, 2005; Chuah & Thai, 2004;among

others). For this study, we collected proxy measures for financial

development andreal sectorand economic growth from theWorld

BanksWorld Development Indicators2009 (WDI) database for the

period from 1980 to 2007. In our analysis, we used GDP per capita

growth rates as a proxy for economic growth (GROWTH). We also

used six variablesto measure financial development andthe size of

the real sector. Our proxy measures forFD incorporate information

from banks and other financial intermediaries in addition to loan

markets.

The first proxy is domestic credit provided by the banking sec-

tor as a percentage of GDP (DCBS). Higher DCBS indicates a higher

degree of dependence upon the banking sector for financing. In

other words, higher DCBS implies higher FD because banks are

more likely to provide the five financial functions discussed in

Levine (1997). It is assumed, however, that banks arenot subject to

mandatedloans to prioritysectors,or obligatedto holdgovernment

securities, which may not be suitable for developing countries.

3 Thetotal number of countriesin theWDI database is 209.However, we dropped

countries that do not have enough data for analysis.

8/12/2019 585-paper-4

4/17


Because of this shortcoming,we also used domestic creditto the

private sector as a percentage of GDP (DCPS) to measure FD. A high

ratio of domestic credit to GDP indicates not only a higher level of

domestic investment, but also higher development of the financial

system. Financial systems that allocate more credit to the pri-

vate sector are more likely to be engaged in researching borrower

firms, exerting corporate control, providing risk management con-

trol, facilitating transactions,and mobilizing savings (Levine, 2005),

which requires a higher degree of financial development.

We also used the broadest definition of money (M3) as a pro-

portion of GDP to measure the liquid liabilities of the banking

system in the economy. We used M3 as a financial depth indicator

because the other two monetary aggregates (M2 or M1) may be

a poor proxy in economies with underdeveloped financial systems

because they are more related to theability of thefinancialsystem

to provide transaction services than to the ability to channel funds

from saversto borrowers (Khan & Senhadji,2000, p. ii93).A higher

liquidity ratio means higher intensity in the banking system. The

assumption here is that the size of the financial sector is positively

associated with financial services (King & Levine, 1993b).4

The fourth indicator of financial development is the ratio of

grossdomesticsavings to GDP (GDS). Pagano (1993) concluded that

the steady state growth rate depends positively on the percent-

age of savings diverted to investment, suggesting that converting

savings toinvestment is onechannel through which financial deep-

ening affects growth. In other words, financial development is

expected to benefit from higher GDS and, consequently, higher

volume of investment. Moreover, in most developing countries,

financial repression and credit controls lead to negative real inter-

est rates that reduce the incentives to save. According to this view,

a higher GDS resulting from a positive real interest rate stimulates

investment and growth (McKinnon, 1973; Shaw, 1973).

We followed the procedure ofLevine et al.(2000) to address the

potential stock-flow problem of our financial variables. The stock-

flowproblem refers to thefact thefinancialbalance sheet items are

measuredat theend ofthe year,whereas GDPis measuredthrough-

outthe year. We deflated end-of-year financial balance sheet itemsby end-of-year consumer price index (CPI), and then we computed

the average of the real financial balance sheet items in yearstand

t1 and divided it by real GDP in year t.5

The fifth and sixth indicators used in this study are the ratio of

tradeto GDP(TRADE)andthe ratioof general government final con-

sumption expenditure to GDP (GOV), respectively. They effectively

measure the size of the real sector and the weight of fiscal policy.

Many developing countries tend to rely heavily on international

trades to achieve economic growth while financial liberalization

is still in progress. In addition, some countries use expansionary

or contractionary fiscal policies for steady economic growth by

adjusting government spending. Finally, we included the inflation

rate (INF) to control for price distortions.

4. Panel estimations and multivariate time-series

methodology

4.1. Panel estimations with convergent term

To examine the general relationship between financial devel-

opment, the real sector, and economic growth, we estimated panel

regressions for each region as well as pooled data. Specifically, to

study the long-term association between GDP per capita and the

4 Nevertheless,M3 maybe influenced by factors other thanfinancial depth, espe-

cially in developed countries.5

SeeAppendix Afor calculation details.

proxy variables, we followed the neo-classical growth model (see

Mankiw, 1995).6 Define growth of real GDP per capita as:

GROWTHi,t= log GDPPCi,t log GDPPCi,t1, i =

1, 2, ...N

(1)

where GDPPC is the real GDP per capita and Nis the number of

countries in the region. Let Qi,0 be the initial level of log(GDPPC)

andQi

the (long-run) steady state GDP per capita. The first-order

approximation of the neo-classical growth model implies thatGROWTHi,t= (Qi,t Q

i,t)

where is a positive convergent parameter. The literature often

implicitly modelsQi

as a linear function of structural parameters;

therefore, a typical growth relationship is:

GROWTHi,t= Qi,t+ Xi,t+ i,t (2)

whereXi,tis a vector of variables controlling for long-run GDP per

capita across countries. Therefore, our regression models are:

GROWTHi,t = 0Qi,1980 +1 FINi,t+2 GDSi,t+ 3 TRADEi,t

+4 GOVi,t+ 5 INFi,t+ i,t (3)

whereQi,1980is the log of GDP per capita and represents the initial

GDP per capita proxy, whereas FINi,t=

DCPSi,t, DCBSi,t, M3i,t

,

represents different proxies for financial depth and development.

In each regression, we included only two financial variables (FIN

and GDS) because DCPS, DCBS and M3 are highly correlated

amongst themselves for most developing countries. Thus, we per-

formedthree separate regressions to studythe impactof financeon

economic growth. To control for business cycles, we calculatednine

non-overlapping-five-year averages for each variable and included

a dummy variable for each quinquennium. We performed ordi-

nary least squares (OLS) regressions using robust-heteroscedastic

errors. Finally, since the number of countries differs in each region,

we used weighted least square regressions (WLS) when estimat-

ing the pooled (worldwide) regression. Each set of regression was

performed on the six geographic regions and on two high-income

groups.

4.2. Multivariate time-series models

The precedent model regressions study association, but not

causality, among variables. To consider dynamic causality, direc-

tion, and timing between financial development and economic

growth, we estimated vector autoregressive (VAR) models (Sims,

1980) andtested whether andwhat proxy variables Granger-cause

economic growth and vice versa. Granger causality tests allow us

to overcome the endogeneity problem presented in panel regres-

sions in the sense that VAR equations consider all variables as

endogenous. In analyzingthe results fromthe VARmodel, we tested

Granger causalityamong variables andfocus on twotools: impulse

response function (IRF) and forecast error variance decomposition(FEVD). IRF shows how one variable responds over time to a single

innovation in itself or in another variable. Innovations in the vari-

ables are represented by shocks to the error terms in the equations

of thestructuralVAR form. More importantly, we computed fore-

cast error variance decompositions of GROWTH to examine what

proxy measures are most important in economic growth over time

and how much they contribute to economic growth.

Our VAR specification includes a total of six variables, including

proxy measures for financialdevelopment (DCPS, andGDS), thereal

sector (TRADE, GOV and INF), and economic growth (GROWTH)

6 Themodel used in this paper hasbeenusedextensively in theliterature. Seefor

example,Barro and Sala-i-Martin (1995),Barro (1997), andBekaert et al. (2005).

8/12/2019 585-paper-4

5/17


across six geographic regions and two high-income groups. For-

mally, thestandardVAR model is expressed as:

Yt= C+

m

s=1

AsYts + et (4)

where Yt is a 61 column vector of 6 variables including proxy

measures (GROWTH, DCPS, GDS, TRADE, GOV, INF); Cand As are,respectively, 61 and 66 matrices of coefficients; m is the lag

length; and etis a 61 column vector of forecast errors. By VAR

construction, the elements of the vector ethave zero means and

constant variances, and are individually serially uncorrelated.7

The ijth component of As measures the direct effect that a

change on the jth variable would have on the ith variable in s

periods.

Weused Todaand Yamamotos (1995) procedure to testGranger

causality. It is well known that F test of causality in VAR is not

valid in the presence of non-stationary series. Toda and Yamamoto

(1995), however, propose a procedure that is robust enough to

address the cointegration features of the series (e.g. it is valid

without regard to the cointegration process of the variables). The

procedurebasically involves four steps.First, findthe highestorder

of integrationin the variables(dmax). Second, findthe optimal num-beroflagfortheVARmodel(m). Third,overfit(on purpose) theVAR

by estimating a (m + dmax) th order VARusing seemingly unrelatedregression (SUR). We used SUR because the Wald test gains effi-

ciency if the VAR is estimated using SUR (Caporale & Pittis, 1999).

Finally, test the null hypothesis of no Granger causality using the

Wald test, which follows a2 distribution withmdegrees of free-dom.

We also used the estimated VAR to calculate impulse response

functions on growth to innovations in each of the variables, as well

as FEDV for each variable. The impulse response functions show

how shock in our financial measures affects growth over time,

whereas the decomposition of forecast error variance provides a

measure of the overall relative importance of the variables in gen-

erating the fluctuations in proxy measures on their own and forother variables.

5. Empirical results

5.1. Summary statistics of proxy measures

Table 1 compares key financial and real indicators along with

the economic growth proxy across geographic regions and income

groups. Among geographic regions made up of developing coun-

tries, Latin America & Caribbean has the highest GDP per capita,

followed by Europe & Central Asia and Middle East & North Africa

(MENA), whereas Sub-Saharan Africa shows the lowest GDP per

capita (see medians).

East Asia & Pacific countries have growth rates comparableto those of high-income countries, reflecting the rapid economic

expansion of many Asian countries in recent decades. Further-

more, Sub-Saharan Africa has the lowest median GDP per capita,

followed by South Asia, which denotes the poverty level of those

regions. Interesting, despite its low GDP per capita, South Asia has

the highest GDP per capita growth among the regions. In general,

developing countries (except those in South Asia) have experi-

enced, at least for one year, negative GDP per capita growth during

7 In the structural VAR system (not shown) the error terms are assumed to be

white-noise disturbances. Under this assumption, errors in the standard form are

individually serially uncorrelated, but there may be contemporaneous correlations

among them. SeeEnders (2009)for a detailed explanation of VAR.

thesample periodmainlydue toeconomicrecession or thepolitical

instability prevalent in those regions.

As expected, high-income OECD countries possess the highest

valuesof DCBS, DCPS, andM3 proxy measures,which representthe

relatively large sizes of their financial systems and their financial

depth. It is obvious that developed countries with efficient finan-

cial intermediaries still tend to rely heavily on domestic credits

provided by the banking sector and have plenty of liquid liabilities

in their banking systems. However, financial depth indicators in

the South Asia and Sub-Saharan Africa regions are relatively low,

implying that they have insufficient credit available to their pri-

vate sectors and inefficient financial systems, which may impair

economic growth in these regions. However, most regions show a

similar size of gross domestic savings as a percentage of GDP.

Latin America & Caribbeanand MiddleEast & North Africacoun-

tries show trade levels similar to those of OECD countries. East

Asia & Pacific has the highest level of trade, whereas South Asia

and Sub-Saharan Africa have the lowest levels. Non-OECD coun-

tries have higher trade levels than OECD countries. In the latter

countries, most trade is in commodities (such as petroleum and

agriculture). Government expenditure is lower for middle- and

low-income countries compared to high-income countries. Europe

& Central Asia and Latin America & Caribbean are the regions with

the highest inflation levels during the period.

5.2. Analysis of panel regressions

Table 2 shows results for panel regressions. Panels A, B, and

C provide different regressions in which domestic credit to the

private sector(DCPS), domestic creditprovidedby thebanking sec-

tor (DCBS) and liquid liabilities (M3), respectively, pair with gross

domestic savings (GDS) as financial development measures. These

financial measures, as well as the other control variables, proxy for

the steady state level of GDP.

The theoretical model explained above suggests that the coeffi-

cient for Q should be negative (see Eq.(2)).As expected, given the

standard results for conditional convergence, the coefficients forQ, when significant, are negative for all regions but Latin America

& Caribbean. These results are consistent with the previous liter-

ature (see, for example,Barro, 1997; Barro & Sala-i-Martin, 1995;

Bekaert et al., 2005)and imply that a low level of initial GDP per

capita is associated with a higher growth rate, conditional on the

other variables. Nevertheless, the significant positive sign in Latin

America & Caribbean suggests a decline in growth of the real GDP

per capita since 1980 in this region.

Panel A displays results when DCPS and GDS are used as prox-

ies for financial development. GDS, when significant, has a positive

sign in all regions, confirming a long-run positive relationship

between savings and growth as predicted inPaganos (1993)the-

oretical model. This is also consistent with the argument that

well-developed domestic financial sectors in developing countriesmay significantly contribute to an increase in savings and invest-

ment rates, which ultimately trigger economic growth (Becsi &

Wang, 1997).GDS is significant in South Asia, Sub-Saharan Africa,

and high-income OECD countries. For example, in South Asia the

GDS coefficient is 2.35 and more than the 2 standard error from

zero. This suggests that on average, a 1% increase in GDS implies a

2.35% increase in growth.

DCPS is significant and positively associated with growth in

East Asia & Pacific and Latin America & Caribbean. The results are

consistent with previous studies, which find a positive relation-

ship between measures of financial development and growth (see

Levine, 2005).

However, the results relating to high-income countries are

surprising, since they indicate a significant negative relationship

8/12/2019 585-paper-4

6/17


Table 1

Summary of statistics by region (19802007).

Economic growth Financial development Real sector

GDP per capita (US $) Growth (%) DCPS (%) DCBS (%) M3 (%) GDS (%) TRADE (%) GOV (%) INF (%)

East Asia & Pacific (N=14)

Mean 1,049.3 3.0 43.0 51.2 51.4 18.1 93.3 14.4 9.9

Median 717.3 2.7 33.4 38.6 41.3 18.2 96.4 14.0 6.9

Max 3,237.2 8.4 128.2 156.8 110.6 38.4 162.6 26.5 34.3Min 297.4 0.0 6.5 6.5 13.7 17.0 40.9 5.0 3.1

Europe & Central Asia (N=20)

Mean 1,842.1 1.0 18.8 30.0 28.3 15.6 85.0 16.2 118.0

Median 1,438.4 1.6 16.3 29.9 26.5 15.8 89.7 17.2 51.3

Max 4,216.6 3.8 44.2 57.4 60.2 32.6 122.9 24.2 414.9

Min 256.0 3.1 6.2 13.2 8.2 0.4 34.2 9.6 12.1

Latin America & Caribbean (N=28)

Mean 2,865.2 1.3 36.6 56.5 46.3 15.5 78.7 14.6 74.8

Median 2,572.1 0.9 33.5 49.4 38.1 15.8 69.4 13.5 15.0

Max 7,149.0 4.0 70.1 177.6 101.3 28.2 180.9 29.8 515.6

Min 547.1 2.5 14.0 20.4 21.8 2.1 20.0 6.9 1.5

Middle East & North Africa (N=12)

Mean 2,026.6 0.9 35.2 58.8 68.5 11.1 72.3 19.1 14.9

Median 1,406.3 1.3 33.7 55.8 60.2 15.3 65.4 16.3 7.8

Max 6,714.0 2.6 70.6 131.9 172.4 34.9 123.8 30.0 77.1

Min 498.8 2.0 5.5 6.7 20.7 23.8 38.4 13.1 4.3

South Asia (N= 7)

Mean 727.3 3.7 21.7 35.6 40.1 20.4 62.4 12.3 7.7

Median 485.8 3.6 22.6 40.2 41.0 13.9 42.8 11.3 7.8

Max 2,403.9 5.9 30.4 50.4 47.8 44.7 163.9 20.6 11.0

Min 193.5 2.1 8.1 6.4 30.3 11.1 22.8 4.7 5.4

Sub-Saharan Africa (N=40)

Mean 849.1 0.5 30.9 80.3 40.7 7.8 71.5 16.7 88.3

Median 298.8 0.4 13.9 23.6 23.8 6.0 60.0 15.0 10.0

Max 5,904.8 5.0 515.0 1702.0 348.5 45.8 160.1 42.5 1302.4

Min 127.9 6.2 1.8 30.2 12.9 38.5 25.4 8.4 2.7

High-income OECD (N=27)

Mean 19,476.5 2.2 85.3 103.6 72.4 23.8 77.7 19.0 5.1

Median 20,251.1 1.9 79.0 97.5 65.5 23.0 68.5 19.0 4.1

Max 36,442.1 5.6 183.1 265.7 194.2 37.8 220.0 27.2 15.3

Min 3,795.0 1.0 39.5 52.4 38.6 12.4 21.8 10.5 1.0

High-income non-OECD (N=20)

Mean 13,098.3 2.3 60.1 62.5 75.3 31.9 143.1 18.8 6.6

Median 10,486.7 2.7 53.7 48.8 64.1 30.5 116.3 19.0 3.9

Max 29,766.1 11.6 146.0 147.5 208.1 64.8 414.7 30.4 47.8Min 2,809.2 2.4 9.7 14.9 11.0 11.9 74.0 8.1 0.6

This table summarizes country-year statistics for six geographic regions and high-income OECD and non-OECD countries classified according to the World Bank. The time-

series averageof eachvariable is calculated andthen statistics arecollected cross-country. Economiesare divided accordingto 2008GNI percapita,calculatedusing theWorld

Bank Atlas method. The groups are: low income, $975 per capita or less; lower middle income, $976 $3855 per capita; upper middle income, $3856$11,905 per capita;

and high income, $11,906 per capita or more. Geographic classifications are assigned only for low-income and middle-income economies. DCPS: domestic credit provided

to private sector; DCBS: domestic credit provided by banking sector; M3: liquid liabilities; GDS: gross domestic savings; TRADE: import plus export; GOV: government

expenditure, all as a proportion of GDP; INF: inflation rate. Detailed variable definitions are presented inAppendix A.

between DCPS and growth. This result contradicts what has been

foundin previous studies andhighlightsthe importanceof studying

the relationship between finance and growth by income groups as

opposed to an aggregation of worldwide economies. In fact, DCPS

is not significant when pooling the worldwide data.Benhabib and

Spiegel (2000)argue that not all indicators of financial develop-ment measure the same forces. We believe that our measures of

financial development are more suitable for developing countries,

since they are weighted more towards financial markets (banking)

than capital market development (stock and bond markets). We

acknowledge that our measures might not be measuring financial

development in the case of high-income OECD countries.8

The worldwide-pooled regression shows that GDS is positive

and significant, implying a long-term association between finance

andgrowth. The level of trade has also positively impacted growth,

8 Nevertheless, we were able to replicateKing and Levines (1993a, 1993b)find-

ing:a significant positiveassociationbetweenDCPS andreal GDPper capita growth

in the period 19601989. DCPS is no longer significant after the 1990s.

whereas government expenditure and inflation have impaired

growth since the 1980s. The latter results are significant for East

Asia & Pacific, Europe & Central Asia, and Sub-Saharan Africa,

denoting that government fiscal policies and price instability have

harmed economic growth in those regions. Also, trade has been an

important driver of growth in East Asia & Pacific, Latin America &Caribbean and Middle East & North Africa.

Panel B describes results when DCBS and GDS serve as proxies

for financial development. As in the previous cases, Q is nega-

tive for East Asia & Pacific and Middle East & North Africa and

positive for Latin America & Caribbean. There are also signifi-

cantlynegative coefficients forGOV in some regions, which implies

that government expenditure has impeded growth. As found in

Panel A, there is a positive relationship between GDS and growth

rate. Moreover, the signs for DCBS enter positively for middle-

and low-income countries but negatively for OECD high-income

countries. These results together suggest a long-term association

between finance and growth. Similar to panel A, trade has posi-

tive association with growth, whereas GOV and INF have negative

association.

8/12/2019 585-paper-4

7/17

8/12/2019 585-paper-4

8/17


Table 3

Forecast error variance decompositions of economic growth in VAR.

Period GOV INF TRADE GDS DCPS GROWTH

East Asia & Pacific

2 years ahead 0.4 12.5 3.5 0.2 6.2 77.3

5 years ahead 0.3 12.8 4.0 0.3 6.2 76.4

10 years ahead 0.4 13.0 4.2 0.4 6.8 75.2

Europe & Central Asia

2 years ahead 2.4 5.4 2.1 2.8 2.0 85.35 years ahead 9.7 4.6 4.8 5.2 3.1 72.7

10 years ahead 15.1 4.6 6.7 5.8 2.8 65.0

Latin America & Caribbean

2 years ahead 0.3 4.7 0.3 0.2 1.8 92.7

5 years ahead 1.3 5.4 1.8 1.5 2.2 87.7

10 years ahead 1.4 5.4 2.2 2.0 2.6 86.4

Middle East & North Africa

2 years ahead 15.3 2.2 9.0 0.4 1.7 71.4

5 years ahead 16.5 2.2 10.9 0.4 1.8 68.2

10 years ahead 17.1 2.5 11.5 0.5 1.7 66.8

South Asia

2 years ahead 4.3 3.1 2.1 14.9 1.1 74.4

5 years ahead 4.3 3.6 2.1 14.9 1.7 73.3

10 years ahead 4.3 3.6 2.1 15.1 2.7 72.2

Sub-Saharan Africa

2 years ahead 0.6 2.9 6.6 0.6 3.5 85.8

5 years ahead 0.7 4.4 7.2 0.6 3.3 83.810 years ahead 0.7 5.1 7.2 0.7 3.4 83.0

High-income OECD

2 years ahead 15.4 13.2 1.7 0.3 1.3 68.1

5 years ahead 15.2 14.9 2.5 0.3 2.5 64.6

10 years ahead 15.1 14.8 2.6 0.3 2.9 64.3

High-income non-OECD

2 years ahead 12.9 0.1 9.8 0.5 13.9 62.8

5 years ahead 13.5 0.2 11.2 0.9 15.9 58.3

10 years ahead 13.9 0.3 11.8 1.0 18.4 54.6

This table summarizes error variance decompositions of economic growth for six geographic regions and two groups of high-income countries classified according to the

World Bank. Geographic classifications are assigned only to low-income and middle-income economies. GROWTH: the difference between natural logarithm of GDP per

capita and its lagged value; DCPS: domestic credit provided to private sector divided by GDP; GDS: gross domestic savings divided by GDP; TRADE: import plus export

divided by GDP; GOV: general government consumption expenditure divided by GDP. All independent variables are in natural logarithm. INF is the log of one plus inflation

rate. The sample period is 19802007.

Panel C portrays results when M3 and GDS serve as proxies forfinancial development. The results are similar to those presented

in panels A and B. That is, the initial GDP per capita is positively

related to growth rate (except in Latin America & the Caribbean),

GDS and TRADE are positively associated with growth rate, and

GOV and INF are negatively related to growth. However, as seen in

thepooled regression as well as theregression forEurope & Central

Asia and high-income OECD countries, M3 is negatively related to

growth rate.

In summary,our results show thattrade has positivelyimpacted

economic growth, whereas government expenditure and inflation

have impaired economic growth worldwide. Also, consistent with

the neo-classical literature, the level of GDS is positively related

with economic growthand both DCPS andDCBSare positively asso-

ciated with economic growth. Thus, given the positive coefficients

forour financialmeasuresin low-and middle-incomecountries, we

can conclude that there is a positive relationship between financial

depth and economic growth in developing countries.

5.3. Analysis of VAR results by geographic regions and income

groups

We turn to theVAR analysis forregions by highlightingthe most

importantresults first, andthen providingdetailedanalysis of each

region. We have decomposed the forecast error of the endogenous

variable GROWTH over different time horizons into components

attributable to unexpected innovations (or shocks) of itself and

proxymeasuresin thedynamicVARsystem.The forecasterrorvari-

ance decompositions of GROWTH in VAR across geographicregions

(and income groups) are presented inTable 3.It is typical in VARanalysis that a variable explains a huge proportion of its forecast

error variance, which is the case in our analysis of GROWTH vari-

ation, which explains the biggest part of itself in all regions. The

second, more important variable in explaining GROWTH variation

is not a finance measure, except in South Asia, where GDS explains

a high proportion of GROWTH variation. Rather, real sector vari-

ables (government expenditure, inflation, or trade) explain more

GDP growth movements in all regions. However, financial depth

still explains an important component of economic growth across

regions.

GROWTH is said to be Granger-caused by proxy measures if

proxy measures help in the prediction of GROWTH, or equivalently

if the coefficients on the lagged proxy measures are statistically

significant. A critical step of the Toda and Yamamoto (1995) proce-

dure is the number of lags in the VAR. Using the Schwartz Bayesian

criterion, the optimal number of lags is four or less for all regions,

and therefore we set this value to four lags.9 We report the results

of the Granger causality tests in Table 4.The first column shows

p-values of the hypothesis that each i variable does not cause

GROWTH, wherei ={DCPS, GDS, TRADE, GOV, INF}. At least either

GDS or DCPS is significant for all regions but be East Asia & the

Pacific and Sub Saharan Africa, meaning that financial develop-

ment Granger-causes economic growth in those regions. TRADE is

9 The maximum order of integration in all series is one. Toda and Yamamoto

(1995) procedurecan be appliedregardlessof whetherthere is cointegrationamong

variables or not.

8/12/2019 585-paper-4

9/17


Table 4

Granger causality test (p-values).

Ho: The variablei ={DCPS, GDS, TRADE, GOV}

does not cause GROWTH

Ho: The variablei = {GROWTH, GDS, TRADE,

GOV}does not cause DCPS

Ho: The variablei = {GROWTH, DCPS, TRADE,

GOV}does not cause GDS

East Asia & Pacific

DCPS 0.94 GROWTH 0.00*** GROWTH 0.02**

GDS 0.31 GDS 0.31 GDS 0.04**

TRADE 0.04** TRADE 0.00*** TRADE 0.43

GOV 0.10 GOV 0.07* GOV 0.64INF 0.09* INF 0.00*** INF 0.40


DCPS 0.00*** GROWTH 0.00*** GROWTH 0.04**

GDS 0.05** GDS 0.06* GDS 0.26

TRADE 0.08* TRADE 0.96 TRADE 0.06*

GOV 0.33 GOV 0.24 GOV 0.25

INF 0.90 INF 0.00*** INF 0.39



GDS 0.03** GDS 0.35 GDS 0.53

TRADE 0.00*** TRADE 0.41 TRADE 0.42

GOV 0.00*** GOV 0.06* GOV 0.77

INF 0.28 INF 0.00*** INF 0.11


DCPS 0.01*** GROWTH 0.00*** GROWTH 0.00***

GDS 0.08* GDS 0.09* GDS 0.02**

TRADE 0.00*** TRADE 0.09* TRADE 0.50GOV 0.11 GOV 0.39 GOV 0.08*

INF 0.00*** INF 0.23 INF 0.30

South Asia

DCPS 0.02** GROWTH 0.17 GROWTH 0.01***

GDS 0.00*** GDS 0.46 GDS 0.97

TRADE 0.02** TRADE 0.00*** TRADE 0.01***

GOV 0.24 GOV 0.11 GOV 0.02**

INF 0.01** INF 0.04** INF 0.02**

Sub-Saharan Africa

DCPS 0.15 GROWTH 0.00*** GROWTH 0.08*

GDS 0.43 GDS 0.25 GDS 0.21

TRADE 0.01** TRADE 0.07* TRADE 0.66

GOV 0.38 GOV 0.25 GOV 0.62

INF 0.00*** INF 0.00*** INF 0.66

High-income OECD countries

DCPS 0.00*** GROWTH 0.00*** GROWTH 0.00***

GDS 0.01** GDS 0.65 GDS 0.49

TRADE 0.11 TRADE 0.64 TRADE 0.30

GOV 0.00*** GOV 0.45 GOV 0.00***

INF 0.00*** INF 0.53 INF 0.05**

High-income non-OECD countries


GDS 0.09* GDS 0.13 GDS 0.43

TRADE 0.04** TRADE 0.02** TRADE 0.07*

GOV 0.52 GOV 0.70 GOV 0.02**

INF 0.01** INF 0.00*** INF 0.00***

In this table, the Todaand Yamamoto(1995) procedure is used to test theGrangercausality amongvariables. This procedure canbe used in presence of cointegration or not.

The table reportsp-values from the WALD test. GROWTH: difference between natural logarithm of GDP per capita and its lagged value; DCPS: domestic credit provided to

privatesector divided by GDP; GDS:grossdomesticsavingsdividedby GDP;TRADE: import plusexportdividedby GDP; GOV:general government consumption expenditure

divided by GDP. All independent variables are in natural logarithm. INF is the log of one plus inflation rate. The sample period is 19802007.

also significant in all regions, implying that trade Granger-causes

EG.10

The second and third columns show Granger causality tests for

DCPS and GDS, respectively. DCPS Granger-causes GROWTH for

all regions but South Asia, while GDS Granger-causes GROWTH

in all regions, implying that FD causes EG. Thus, Granger causal-

ity tests imply a two-way causality between finance and growth

in all regions but Sub-Saharan Africa and East Asia & Pacific,

10 Note that the emphasis in Granger causality tests is on short-runrelationships,

because the results of panel regression and cointegration tests strongly suggest the

presenceoflong-run linkagesbetween financialdevelopment and economicgrowth.(Theresultsof cointegrationtests arenot shown to preservespace andare available

upon request.)

where the direction is from finance to growth. Our results, for

most of the regions, are consistent with the findings of Shan

et al. (2001), and Demetriades and Hussein (1996), who found

bi-directional causality between finance and growth, and con-

trary to Christopoulos and Tsionas (2004), who found that the

direction is from finance to growth. Moreover, our results give

some support to the theoretical models of Blackburn and Huang

(1998) and Khan (2001), which predict two-way causality between

finance and growth. However, causality runs from growth to

finance in South Asia and Sub-Saharan Africa. This result sup-

ports the view of Gurley and Shaw (1967), Goldsmith (1969),

and Jung (1986), who hypothesize that in developing countries,

growth leads finance because of the increasing demand forfinancial

services.

8/12/2019 585-paper-4

10/17


Panel A. East Asia & Pacific

Growth response to GDS shock Growth response to DCPS shock

Panel B. Europe & Central Asia


Panel C. Latin America & Caribbean


-1.20

-1.00

-0.80

-0.60

-0.40

-0.20

0.00

0.20

0.40

0.60

0.80

1.00

1.20

1 2 3 4 5 6 7 8 9 1 0 11 12 13 14 15

-1.20

-1.00

-0.80

-0.60

-0.40

-0.20

0.00

0.20

0.40

0.60

0.80

1.00

1.20

1 2 3 4 5 6 7 8 9 1 0 11 12 13 14 15

-1.20

-1.00

-0.80

-0.60

-0.40

-0.20

0.00

0.20

0.40

0.60

0.80

1.00

1.20

1 2 3 4 5 6 7 8 9 1 0 11 12 13 14 15

-1.20

-1.00

-0.80

-0.60

-0.40

-0.20

0.00

0.20

0.40

0.60

0.80

1.00

1.20

1 2 3 4 5 6 7 8 9 1 0 11 12 13 14 15

-1.20

-1.00

-0.80

-0.60

-0.40

-0.20

0.00

0.20

0.40

0.60

0.80

1.00

1.20

1 2 3 4 5 6 7 8 9 1 0 11 12 13 14 15

-1.20

-1.00

-0.80

-0.60

-0.40

-0.20

0.00

0.20

0.40

0.60

0.80

1.00

1.20

1 2 3 4 5 6 7 8 9 1 0 11 12 13 14 15

Panel D. Middle East & North Africa


-1.20

-1.00

-0.80

-0.60

-0.40

-0.20

0.00

0.20

0.40

0.60

0.80

1.00

1.20

1 2 3 4 5 6 7 8 9 1 0 11 12 13 14 15

-1.20

-1.00

-0.80

-0.60

-0.40

-0.20

0.00

0.20

0.40

0.60

0.80

1.00

1.20

1 2 3 4 5 6 7 8 9 1 0 11 12 13 14 15

Fig. 1. Generalized impulse response functions of growth. This figure showsPesaran and Shins (1998)generalized impulse response functions of GROWTH to a shock in

GDS and DCPS, respectively. A generalized impulse response function is invariant to variable ordering. GROWTH: difference between natural logarithm of GDP per capita

and its lagged value; DCPS: domestic credit provided to the private sector; GDS: gross domestic savings. The horizontal axis is the number of years following the shock and

the vertical axis is the percent growth rate of GDP per capita (difference in log). The vertical axis has the same scale across regions. The sample period is 19802007.

8/12/2019 585-paper-4

11/17


Panel D. South Asia


Panel E. Sub-Saharan Africa


-1.20

-1.00

-0.80

-0.60

-0.40

-0.20

0.00

0.20

0.40

0.60

0.80

1.00

1.20

151413121110987654321

-1.20

-1.00

-0.80

-0.60

-0.40

-0.20

0.00

0.20

0.40

0.60

0.80

1.00

1.20

151413121110987654321

-1.20

-1.00

-0.80

-0.60

-0.40

-0.20

0.00

0.20

0.40

0.60

0.80

1.00

1.20

151413121110987654321

-1.20

-1.00

-0.80

-0.60

-0.40

-0.20

0.00

0.20

0.40

0.60

0.80

1.00

1.20

151413121110987654321

Panel F. High-Income OECD Countries


Panel G. High-Income Non- OECD Countries


-1.20

-1.00

-0.80

-0.60

-0.40

-0.20

0.00

0.20

0.40

0.60

0.80

1.00

1.20

151413121110987654321

-1.20

-1.00

-0.80

-0.60

-0.40

-0.20

0.00

0.20

0.40

0.60

0.80

1.00

1.20

151413121110987654321

-1.20

-1.00

-0.80

-0.60

-0.40

-0.20

0.00

0.20

0.40

0.60

0.80

1.00

1.20

151413121110987654321 151413121110987654321

Fig. 1. (Continued).

Since our goal is to assess the role of the financial sector in

economic growth, we also investigate the dynamic relationships

among proxy measures and how two measures of financial devel-

opment (GDS and DCPS) affect economic growth (GROWTH) over

time.11 Choleski decomposition is generally used to identify the

system of equations in order to get the impulse response func-

11 To save space, we have concentrated on the impact of shock on finance on

growth. Thus, we have not reported the IRF of our financial measures to shocks

in growth.

tion. However, this decomposition implies that the ordering of

variables matters; in other words, different ordering mayyield dif-

ferent results. Therefore, we use the generalized impulse response

function proposed byPesaran and Shin (1998),which is invariant

to the ordering of the equations. Fig. 1illustrates how GROWTH

responds over time to shock innovation in DCPS and GDS, respec-

tively, by regions. We use the same scale in the axis to assess the

magnitude of the shock on growth among regions.

A positive shock on GDS causes GROWTH to increase in the

first few years for most of the regions. The highest jumps in GDS

magnitude occur in East Asia & Pacific, Europe & Central Asia, and

8/12/2019 585-paper-4

12/17


high-income non-OECD countries. On the other hand, DCPS has

a negative effect on growth during the first two years but turns

positive in the long-run for most countries. The only exception

is for the high-income non-OECD countries, where DCPS yields

a negative effect on GROWTH. Also, the highest impact of DCPS

shock on GROWTH is seen in the Middle East & North Africa region,

where the response is negative growth but a later jump to positive

growth. In summary, savings appears to be an important finance

variable in determining growth in developing countries, whereas

DCPS encompasses marginal effect on growth. Thus, the results are

consistent with the assertion that well-developed financial sectors

may help to increase savings and therefore investment, which in

turn is translated into economic growth because of the increased

investment.

In next sub-sections, we analyze and derive some policy

implications for each region given the results of the VAR anal-

ysis. Therefore, we will refer to error variance decomposition of

growth, Granger causality tests between finance and growth, and

the impulse response function of growth to shocks in finance

(Tables 3 and 4,andFig. 1,respectively) together for each region.

5.3.1. East Asia & Pacific (low- and middle-income countries)

DCPS explains 6.8% of variation in growth rate after 10 years

in East Asia & Pacific countries. Furthermore, Fig. 1 shows that

a shock in DCPS causes growth to decrease in the short-run (the

highest decrease for middle- and low-income countries) and later

to increase to a positive long-run growth rate. This long-term rela-

tionship is significant (as shown previously in Panel A ofTable 2).

However, DCPS does not Granger-cause growth in the short-run

(see Granger causality test), but growth Granger-causes DCPS,

implying one-way causality.

On theotherhand, GDSonlyexplains0.4%of variationin growth

rate after 10 years.However, there is no significant Granger causal-

ity from GDS to GROWTH, but there is a significant causality from

GROWTH to GDS. It seems that policies designed to increase GDS

and DCPS have not had significant effects in East Asia & Pacific.

Rather, theregion hasenjoyed more growthbecause of trading and,thus,policiesfocused ontrading mighthave more benefitthanpoli-

cies designed to increase DCPS and GDS. The increased trade might

continue to foster production and, therefore, economic growth,

which might implicitly help financial development as suggested

by the Granger causality test. However, inflation accounts for 13%

of growth variation and is significantly and negatively related to

growth, suggesting that inflation in the regions has impaired eco-

nomic growth, and, therefore, that high inflationary policies should

be avoided.

5.3.2. Europe & Central Asia (low- and middle-income countries)

GDS accounts for 5.8% of variation in growth after 10 years in

this region. As seen in the impulse response function, GDS will

cause growth to increase and there is a significant Granger causal-ity from GDS to GROWTH and from GROWTH to GDS, implying

two-way direction. DCPS and GROWTH causality is significant and

bi-directional as well. The impulse response function shows that a

shock in GDSwillcauseGROWTH toincrease.In summary, financial

development hassomewhathelped economic growth in theregion

and accordingly, the region may benefit from policies designed to

improve the financial system.

5.3.3. Latin America & Caribbean (low- and middle-income

countries)

INF explains the second-highest proportion of growth variation.

DCPS and GDS explain only 2.6% and 2.0%, respectively. However,

DCPS is significant in the long run and the impulse response func-

tion shows that innovations to DCPS cause a short-term decline in

growth that gradually ends in a rise in growth in the long term. A

shock in GDS causes growth in the short term but it gradually dis-

appears in the long term. However, there is no evidence that DCPS

Granger-causes growth, but GDS does, implying two-way causal-

ity. It seems that the region should pay more attention to the level

of government expenditure (fiscal policies) and avoid inflationary

policies. Trade is also an important variable to explain growth;

hence, policies focused on improvement of trading might lead to

economic growth in the region.

5.3.4. Middle East & North Africa (low- and middle-income

countries)

DCPS and GDS explain only a small proportion of the varia-

tion compared with the real sector (1.7% and 0.5%, respectively)

in this region. Nevertheless, a DCPS shock causes the growth rate

to rapidly increase and then die out after four years (see Fig. 1).

As a matter of fact, DCPS Granger-causes GROWTH and GROWTH

Granger-causes DCPS (bi-directional causality). However, TRADE

explains a higher proportion of growth variation and it Granger-

causes GROWTH, implying that trading is a critical variable in the

region. The results indicate that efforts to reform and deepen the

financial system in the Middle East & North Africa region would

prove fruitful onlyif accompanied by policies thatprovidean incen-tive to develop trade.

5.3.5. South Asia (low- and middle-income countries)

GDS explains 15.1% of GROWTH variation in this region, which

is significantly higher than other geographic regions. Moreover, a

shock in GDS causes growth rate to increase promptly (seeFig. 1).

GDS Granger-causes GROWTH and GROWTH Granger-causes GDS,

thus implying a two-way causality between finance and growth.

Given the results in the regression above, and the causality test, it

seems that GDS has been more important than DCPS for financial

policy purposes.

5.3.6. Sub-Saharan Africa (low- and middle-income countries)

The financial variables explain only a very low proportion ofvariation of GDP per capita growth. The impulse response function

also shows that shocks of these variables have insignificant effects

on growth. A shock in GDScauses GROWTHto increase butthisdies

out quickly, whereas a shock in DCPS causes GROWTH to decline,

but it recovers one period later. Granger causality tests indicate

one-way causality from GROWTH to financial measures.

The only variable that explains a significant proportion of

growth variation is TRADE. It seems, therefore, that the Sub-

Saharan Africa region should increase trading to enhance growth.

However, from the regression above, the Granger causality test

and impulse response functions suggest thatrising domestic saving

may boost growth. Since this region is the poorest in our sample,

it is not surprising that DCPS and GDS have little effect on growth,

given the lowlevel of these variables compared with other regions.Therefore, policies directed at increasing these variables should

attract a higherlevel of investments to enhance long-run economic

growth.

5.3.7. High-income OECD countries

The variance decomposition implies that proxy measures for

the real sector play a more important role in explaining GROWTH

fluctuations compared to those of financial development for OECD

countries. GOV and INF shocks explain 15.1% and 14.8% of growth

fluctuations, respectively, whereas DCPS and GDS shocks explain

2.9% and 0.3% respectively. A shock to GDS produces an increase

in GROWTH that quickly dies out. Furthermore, there is two-way

causalitybetween finance (DCPSand GDS)and GROWTH. However,

considering the results of the regression, variance decomposition,

8/12/2019 585-paper-4

13/17


causalitytest, andimpulse response function,it seems that domes-

tic credit and gross domestic savings have not been as important

for growth as trade, government fiscal policies, and inflation.

5.3.8. High-income non-OECD countries

DCPS explains 18.4% of growth variation in non-OECD coun-

tries,and this proportion is significantly higherthan the proportion

observed in any other region. However, as shown in the regres-sion and impulse response function,the relationship between DCPS

and GROWTH is negative. The impulse response function shows

that a positive shock in DCPS would cause the steepest decline in

GROWTH compared to the other groups, culminating in a long-

run decline. The Granger causality test shows that the direction

of the causality is from GROWTH to DCPS. However, there is a

positive two-way causality between GDS and GROWTH. Addition-

ally, there is two-way causality between trade and GROWTH, most

likely because countries in this group are basically exporters of

commodities (mainly petroleum).

6. Conclusions

We examined panel regressions with cross-sectional coun-

tries and time-series proxy measures to study linkages between

financial development and economic growth in low, middle and

high-income countries as classified by the World Bank. We also

performed various multivariate time-series models in the frame

of VAR analysis, forecast error variance decompositions, impulse

response functions, and Granger causality tests to document the

direction and relationship between finance and growth in these

countries with the objective of documenting the progress in finan-

cial liberalization and exploring some policy implications.

Consistent withBekaert et al. (2005)andBarro (1997),among

others, we found that a low initial GDP per capita level is asso-

ciated with a higher growth rate, after controlling for financial

and real sector variables. Furthermore, in agreement with King

and Levine (1993a), and Levine et al. (2000), among others, we

foundstrong long-run linkages between financial development andeconomic growth. Specifically, as predicted in neo-classical mod-

els (Pagano, 1993),domestic gross savings is positively related to

growth. We also found that domestic credit to the private sector is

positively related to growth in East Asia & Pacific, and Latin Amer-

ica & Caribbean, butis negatively related to growth in high-income

countries.

Using Granger causality tests to study the direction between

finance and growth, we found that, in the short run, there is

two-way causality between finance and growth in all regions but

Sub-Saharan and East Asia & Pacific. This result is consistent with

the findings ofShan et al. (2001), and Demetriades and Hussein

(1996), who found bi-directional causality between finance and

growth, and contrary to Christopoulos and Tsionas (2004), who

found that the direction is from finance to growth. Moreover, ourresults give some support to the theoretical models ofBlackburn

and Huang (1998) and Khan (2001), which predict a two-way

causality between finance and growth.

However, Sub-Saharan Africa and East Asia & Pacific have

causality that runs from growth to finance, supporting the view of

Gurley and Shaw (1967),Goldsmith (1969),andJung (1986),who

hypothesized that in developing countries, growth leads finance

because of the increasing demand for financial services. These

two regions have the lowest GDP per capita in our sample, and

not surprisingly, their underdeveloped financial systems do not

Granger-cause growth. However, there is a long-term association

between finance and growth, as shown in the regression. Pol-

icy should be centered on improving international trade in these

regions with the objective of fostering growth.

Ourempirical results based on Granger causalitytestsand panel

regressions do notattempt to answer the question What will hap-

pen in the future? Rather, they tell us what has happened in the

past. More specifically, the question of whether finance leads to

growth will still be subject to debate. We found that there has

been a positive association between finance and economic growth

for developing countries but contradictory results for high-income

countries.

Nevertheless, given the evidence in our empirical analysis for

middle- and low-income countries, it seems that well-functioning

financial systems may boost economic growth in these countries.

Of course, we have documented this positive relationship, but

development of financial systems in developing countries is not

a panacea since other real variables such as trade and government

fiscal policies are important determinants of growth. Rather, pol-

icy makers and international organizations such the International

Monetary Fund or theWorldBank shouldconsidera countryslegal

system, political stabilities, and stage of financial development

when designing policies to boost economic growth and reduce

poverty. In summary, while financial development may be neces-

sary, it is not sufficient to attain a steady economic growth rate in

developing countries.

Appendix A. Time-series averages of variables by country

(19802007)

This appendix summarizes time-series statistics for six geo-

graphic regions and high-income OECD and non-OECD countries

classified according to the World Bank. The time-series average of

each variable is calculated, and then statistics are collected cross-

country. Economies are divided according to 2008 GNI per capita,

calculated using the World Bank Atlas method. The groupsare: low

income, $975per capita or less; lowermiddle income, $976$3,855

per capita; upper middle income, $3,856$11,905 per capita; and

high income, $11,906per capita or more.Geographic classifications

are assigned only for low- and middle-income economies.

Balance sheet financial variables are adjusted to address the

potential stock-flow problem (Levine et al., 2000). Our measures

of financial variables are calculated as follows:

FINi,t=12 [FINi,t1/CPIt1] + FINi,t/CPIt

GDPt

where FINi ={DCPS, DCBS, M3, GDS}.

DCPS: Domestic credit provided by the banking sector, which

includes all credit to various sectors on a gross basis, with

theexception of creditto the central government, which is

net. The banking sector includes monetaryauthorities and

deposit money banks,as well as other banking institutions

wheredata are available(includinginstitutions thatdo notaccepttransferable deposits butdo incur such liabilities as

time and savings deposits) (WDI, 2009).

DCBS: Domestic credit provided by the banking sector, which

covers claims on private non-financial corporations,

households, and non-profit institutions (WDI, 2009).

M3: Liquid liabilities, the broadest definition of money.

GDS: Gross domestic savings, which are calculated as GDP

less final consumption expenditure (formerly total con-

sumption). Final consumption expenditures cover the

consumption expenditures by households and the general

government (WDI, 2009).

TRADE: Import plus export.

GOV: Government expenditure.

INF: Inflation rate.

8/12/2019 585-paper-4

14/17


Table A

Time-series average of variables (19802007).



East Asia & Pacific

Cambodia 313.2 6.1 6.5 6.5 13.7 4.9 102.1 5.0 4.3

China 697.8 8.4 86.5 93.2 92.4 38.4 40.9 14.8 6.4

Fiji 1,944.9 0.7 31.4 38.0 44.2 14.5 112.3 17.0 4.6Indonesia 698.3 3.5 30.9 39.2 38.4 29.8 55.1 8.7 10.9

Lao PDR 297.4 3.4 7.2 8.5 14.1 12.3 55.3 7.9 26.4

Malaysia 3,237.2 3.6 128.2 156.8 110.6 36.6 162.6 13.3 3.1

Mongolia 484.0 1.9 16.2 20.8 29.0 19.2 112.3 19.7 34.3

Papua New Guinea 657.0 0.0 19.4 27.9 33.2 23.1 106.4 20.0 7.3

Philippines 964.4 0.8 35.3 51.0 47.2 17.1 76.2 10.2 9.6

Solomon Islands 736.3 1.3 22.3 33.1 28.3 14.9 119.6 22.1 10.6

Thailand 1,687.0 4.5 91.6 109.0 83.2 30.3 88.6 11.2 4.0

Tonga 1,408.8 1.8 51.6 57.2 36.3 17.0 82.8 17.3 7.6

Vanuatu 1,221.8 0.8 36.0 34.8 101.4 8.7 101.1 26.5 4.9

Vietnam 341.9 4.9 38.7 40.9 47.2 20.3 91.8 7.3 4.8


Albania 1,105.2 1.7 8.0 47.9 60.2 7.2 52.0 11.0 26.9

Armenia 732.2 3.6 8.9 13.2 18.7 0.4 80.9 12.5 391.8

Azerbaijan 916.7 1.7 6.2 19.0 20.8 21.3 93.6 15.2 261.5

Belarus 1,386.2 2.7 11.7 21.3 18.5 23.6 122.9 20.2 329.9

Bulgaria 1,654.6 2.2 37.6 55.7 55.9 20.7 98.0 16.6 88.8

Croatia 4,216.6 1.5 44.2 57.4 49.0 14.5 103.2 24.2 212.3

Georgia 1,151.7 1.2 9.3 18.0 11.5 13.8 80.6 11.7 21.8

Kazakhstan 1,454.6 2.2 20.6 17.5 18.3 24.4 90.9 12.2 156.7

Kyrgyz Republic 328.5 0.6 6.4 14.0 16.6 5.6 88.6 19.5 13.2

Latvia 3,597.3 2.6 31.4 34.5 31.2 26.1 101.3 15.3 30.3

Lithuania 3,691.2 1.7 21.7 22.4 27.0 16.2 107.6 18.9 38.4

Macedonia, FYR 1,767.9 0.0 25.4 29.8 26.5 7.1 93.4 20.0 12.1

Moldova 552.4 1.1 13.7 32.0 30.2 9.8 114.6 17.8 15.4

Poland 4,139.8 3.8 24.3 35.5 38.4 18.6 58.5 19.6 51.0

Romania 1,924.9 1.3 13.7 44.8 36.3 17.4 64.6 11.0 77.3

Russian Federation 2,070.3 0.3 17.2 27.2 25.3 32.6 56.8 17.8 111.3

Serbia 1,422.2 1.5 27.4 30.0 19.7 0.3 61.6 19.8 39.6

Tajikistan 256.0 3.1 15.8 18.0 8.2 12.4 110.0 11.6 14.9

Turkey 3,527.0 2.6 16.8 31.8 26.6 15.5 34.2 9.6 51.6

Ukraine 946.1 1.1 14.8 30.0 27.8 26.1 87.5 19.5 414.9


Algeria 1,871.3 0.5 30.5 53.4 60.4 34.9 54.4 15.9 10.6

Djibouti 869.0 2.0 35.2 39.9 67.5 1.4 99.8 30.0 4.4Egypt, Arab Rep. 1,285.5 2.6 40.9 97.5 88.0 14.5 53.3 13.1 11.2

Iran, Islamic Rep. 1,527.1 1.4 32.2 58.1 48.0 29.5 38.4 14.9 19.6

Jordan 1,872.0 0.6 70.6 91.8 108.6 2.1 123.8 24.6 5.0

Lebanon 4,298.1 1.3 67.8 131.9 172.4 12.1 71.2 16.5 77.1

Libya 6,714.0 1.6 20.2 11.2 40.5 25.8 55.5 23.2 4.4

Morocco 1,259.8 1.8 43.1 68.6 65.6 18.7 57.9 17.3 4.6

Syrian Arab Republic 1,113.2 0.9 9.0 53.2 59.9 16.2 59.5 15.6 11.8

Tunisia 1,763.3 2.5 61.2 67.4 51.8 21.8 88.7 16.0 4.9

West Bank and Gaza 1,246.5 1.3 6.1 6.7 20.7 23.8 87.0 25.6 4.3

Yemen, Rep. 498.8 1.3 5.5 25.0 39.0 10.8 78.5 16.1 20.5


Argentina 7,149.0 0.8 20.5 38.0 23.8 20.6 22.8 10.2 302.1

Belize 2,744.9 2.4 43.2 50.5 44.6 13.8 117.4 16.1 2.9

Bolivia 962.7 0.2 36.4 42.0 36.5 12.4 50.9 13.3 515.6

Brazil 3,568.8 0.7 50.5 84.3 44.7 20.1 20.0 16.3 432.7

Chile 3,891.5 3.3 66.1 83.7 37.2 23.5 59.3 11.7 11.9

Colombia 2,280.5 1.7 30.3 38.8 30.4 17.9 33.6 14.1 17.8Costa Rica 3,557.7 1.8 20.3 31.6 28.4 16.7 79.7 13.7 18.9

Dominica 3,122.3 3.0 47.7 59.5 66.4 10.8 114.0 21.4 3.0

Dominican Republic 1,866.4 2.5 40.4 48.2 34.0 15.3 72.5 6.9 17.2

Ecuador 1,364.0 0.7 23.1 25.4 21.8 19.7 56.4 12.7 32.7

El Salvador 1,886.7 0.8 34.0 41.5 39.1 3.4 58.4 11.3 11.2

Grenada 3,049.4 2.9 56.6 69.2 80.2 12.1 113.6 18.3 3.5

Guatemala 1,593.7 0.4 19.1 29.7 26.6 8.3 46.6 7.3 11.6

Guyana 824.7 1.0 40.4 177.6 85.3 15.9 180.9 22.0 6.6

Haiti 547.1 2.5 14.0 34.0 34.6 4.1 43.3 8.8 14.8

Honduras 1,129.0 0.9 33.0 36.8 35.8 16.6 88.6 12.9 13.1

Jamaica 2,872.9 1.0 25.6 56.1 54.7 16.7 99.8 15.2 18.0

Mexico 5,381.8 0.9 19.1 42.4 27.2 22.7 44.2 10.1 33.7

Nicaragua 796.0 0.4 30.1 105.7 40.2 2.1 66.3 19.7 7.6

Panama 3,617.7 1.8 70.1 72.9 58.7 26.5 155.0 16.0 1.5

Paraguay 1,394.0 0.0 21.3 23.5 24.9 15.6 79.0 8.9 15.3

Peru 2,083.5 0.7 18.0 20.4 25.4 21.2 34.5 9.8 475.9

St. Kitts and Nevis 5,815.1 4.0 60.0 95.3 101.3 18.8 129.4 19.9 3.4

8/12/2019 585-paper-4

15/17


Table A (Continued)



St. Lucia 3,497.1 3.0 63.8 68.6 72.3 1 3.9 137.8 18.0 3.4

St. Vincent and the Grenadines 2,399.3 3.7 46.7 59.7 73.9 11.7 131.8 21.0 3.3

Suriname 2,223.4 0.4 25.9 60.0 63.0 9.7 74.6 29.8 46.4

Uruguay 5,566.4 1.3 40.2 53.4 47.5 15.6 43.6 12.7 39.7

Venezuela, RB 5,038.6 0.0 28.3 32.6 36.8 28.2 49.8 10.9 30.9South Asia

Bangladesh 298.7 2.4 29.7 30.3 12.4 4.7 6.0

Bhutan 626.2 5.9 6.4 32.6 25.3 19.0 7.9

India 381.5 4.1 50.4 47.8 22.6 11.3 7.8

Maldives 2,403.9 5.4 43.5 47.7 44.7 20.6 5.4

Nepal 193.5 2.1 31.5 37.9 11.1 8.8 8.4

Pakistan 485.8 2.5 47.6 43.1 12.7 11.4 7.5

Sri Lanka 701.8 3.6 40.2 41.0 13.9 10.4 11.0

Sub-Saharan Africa

Angola 748.6 2.2 5.4 20.0 24.3 36.1 640.2

Benin 298.8 0.4 16.2 25.2 1.9 11.9 6.1

Botswana 2,772.0 5.0 30.2 25.1 40.4 24.2 10.0

Burkina Faso 199.8 1.8 11.3 18.9 3.5 19.9 3.7

Burundi 127.9 1.1 27.0 20.1 3.8 15.7 10.4

Cameroon 716.3 0.0 20.2 18.0 20.3 10.2 5.7

Cape Verde 984.0 3.2 54.3 62.9 5.4 16.5 4.9

Central African Rep. 259.7 1.2 14.9 17.3 1.7 13.7 3.4

Chad 187.6 2.2 12.3 12.9 1.6 8.6 3.9

Congo, Dem. Rep. 156.8 3.6 8.7 13.2 8.5 8.9 1302

Congo, Rep. 1,121.1 0.5 18.4 16.2 36.6 17.0 4.5

Ethiopia 130.7 0.7 36.0 30.5 9.0 11.4 6.6

Gabon 4,673.0 0.6 19.1 17.5 45.8 14.0 3.7

Gambia, The 310.5 0.3 23.6 29.9 7.6 19.7 10.5

Ghana 230.1 1.0 23.8 20.3 6.1 10.7 31.8

Guinea-Bissau 163.1 0.4 16.8 28.7 0.8 14.0 26.9

Kenya 422.9 0.2 44.2 41.4 14.2 17.0 13.0

Lesotho 388.2 2.5 10.5 35.4 38.5 23.5 11.2

Liberia 298.7 6.2 324.1 125.9 4.1 17.0 4.9

Madagascar 253.3 1.2 23.8 22.3 5.4 8.4 15.8

Malawi 142.7 0.2 23.7 21.9 6.4 15.8 21.7

Mali 235.6 0.5 18.5 22.9 5.9 11.6 3.0

Mauritania 432.2 0.2 26.5 19.8 1.9 22.8 6.7

Mauritius 2,975.0 4.1 68.5 74.1 21.8 13.4 7.1

Mozambique 214.2 2.0 1702 348.5 0.5 11.3 26.4

Namibia 1,837.8 0.4 48.5 40.3 13.6 28.5 5.1Niger 188.6 1.7 13.4 14.4 4.9 13.0 3.1

Rwanda 237.2 0.1 12.5 15.6 0.7 12.2 6.6

Senegal 454.7 0.4 29.9 24.7 6.3 15.7 4.5

Seychelles 5,904.8 1.8 75.1 65.2 20.4 29.3 2.7

Sierra Leone 224.6 0.7 39.4 18.3 3.7 10.6 41.8

South Africa 3,199.2 0.3 102.8 128.1 50.5 21.9 18.5 10.1

Sudan 335.2 2.3 6.6 52.7 19.1 8.9 9.6 44.5

Swaziland 1,243.8 2.3 17.5 14.7 25.3 5.2 18.5 10.7

Tanzania 277.0 1.7 8.4 17.9 21.6 5.4 13.9 20.1

Togo 261.2 1.4 20.6 22.8 32.7 7.5 14.3 4.7

Uganda 220.1 2.3 5.7 12.6 13.4 5.0 11.9 44.3

Zambia 365.3 0.6 11.0 54.8 24.7 12.2 17.9 53.9

Zimbabwe 596.0 1.4 24.0 45.6 41.0 14.7 18.8 1007

High-income non-OECD

Antigua and Barbuda 7,449.7 4.0 53.4 69.1 72.5 28.7 19.8 4.6

Bahamas, The 15,391.0 0.7 57.0 69.1 57.9 22.9 13.6 3.5

Bahrain 11,024.7 1.0 49.8 30.8 66.1 34.7 20.1 1.0Barbados 7,873.4 0.4 54.3 70.0 62.1 17.7 19.1 4.0

Brunei Darussalam 19,780.4 1.9 46.8 27.0 69.0 41.4 21.5 1.9

Cyprus 9,696.4 3.4 128.4 147.5 146.8 18.9 16.2 3.9

Equatorial Guinea 2,809.2 11.6 9.7 14.9 11.0 42.9 17.4 4.0

Estonia 4,070.6 2.8 39.8 37.9 38.1 24.0 19.2 16.3

Hong Kong, China 21,847.0 3.9 146.0 137.3 208.1 31.6 8.1 4.7

Israel 16,621.5 1.9 69.2 106.1 82.1 11.9 30.4 47.8

Kuwait 16,773.2 0.8 61.3 79.5 77.6 30.3 26.8 3.0

Macao, China 14,402.8 5.1 69.4 52.7 138.9 50.4 9.5 3.6

Malta 7,589.2 3.2 82.9 95.7 109.0 16.7 19.0 2.5

Oman 7,503.2 3.3 26.5 23.8 28.7 32.1 24.7 1.6

Qatar 29,766.1 2.6 29.6 36.7 41.3 64.8 22.8 4.0

Saudi Arabia 9,948.7 1.7 54.0 42.5 43.5 30.7 27.0 0.6

Singapore 17,633.2 4.3 100.6 80.1 109.2 44.0 10.6 1.7

Slovenia 9,631.7 2.8 36.3 42.7 41.0 24.3 18.9 9.5

Trinidad and Tobago 6,338.5 1.6 43.1 44.9 51.9 28.2 14.4 7.6

8/12/2019 585-paper-4

16/17


Table A (Continued)



United Arab Emirates 25,815.4 2.4 43.6 42.4 51.9 41.4 16.8

High-income OECD

Australia 18,363.1 1.9 67.1 77.2 58.9 23.5 18.4 4.6

Austria 20,838.1 1.9 90.4 113.8 75.1 23.6 18.8 2.7

Belgium 19,753.3 1.7 54.4 97.5 22.7 22.1 3.0Canada 20,704.7 1.7 107.7 125.1 84.8 22.5 21.0 3.5

Czech Re

Documents

585-paper-4